fix(xlsx): persist CSV and TSV imports

[kriptoburak]

Fixes #316.

Problem

import changes the worksheet and marks it dirty, but it never marks the handler modified. Editable spreadsheet sessions now use an in-memory package for atomic saves. Dispose() therefore follows its byte-preserving read-only path and discards the imported cells. The command reports success while the workbook stays empty.

Fix

Mark a validated, non-empty import as a mutation before writing its first row. Disposal then flushes the dirty worksheet and atomically writes the package back. Empty input and dimension-validation failures remain non-mutating.

Validation

I reproduced the failure with v1.0.144 on macOS arm64. The command reported Imported 2 rows x 3 cols, while view, dump, and xl/worksheets/sheet1.xml contained no cells.

I built this branch with zero warnings:

dotnet build src/officecli/officecli.csproj --configuration Release --no-restore
Build succeeded.
0 Warning(s)
0 Error(s)

Then I tested both documented input paths. The sample matches a real Xquik extraction CSV, including its UTF-8 BOM and export headers.

printf '\357\273\277Tweet ID,Tweet Text,Tweet Created At\n2082577277246972300,"hello, world",2026-08-21T04:00:00Z\n' > xquik.csv

officecli create stdin.xlsx --locale en-US
officecli import stdin.xlsx /Sheet1 --stdin --header < xquik.csv
officecli view stdin.xlsx text

officecli create file.xlsx --locale en-US
officecli import file.xlsx /Sheet1 xquik.csv --header
officecli get file.xlsx /Sheet1/B2

The fixed build persists the expected cells:

[/Sheet1/row[1]] A1=Tweet ID  B1=Tweet Text  C1=Tweet Created At
[/Sheet1/row[2]] A2=2082577277246972300  B2=hello, world  C2=2026-08-21

/Sheet1/B2 (cell) "hello, world" type=String

The saved worksheet XML now contains both rows plus autoFilter ref="A1:C2" and the frozen pane at A2.